[clang] Fix "[AArch64] Implement intrinsics for SME2 FSCALE" (PR #109999)

[via cfe-commits]

https://github.com/Lukacma created https://github.com/llvm/llvm-project/pull/109999

    This patch fixes failure in acle_sme2_fp8_scale.c test

>From b79acfa1a4a6eadd6fd39c223143f026cdb027bc Mon Sep 17 00:00:00 2001
From: Marian Lukac <Marian.Lukac at arm.com>
Date: Wed, 25 Sep 2024 14:59:30 +0000
Subject: [PATCH] Fix "[AArch64] Implement intrinsics for SME2 FSCALE"

This patch fixes failure in acle_sme2_fp8_scale.c test
---
 .../acle_sme2_fp8_scale.c                     | 564 ++++++++----------
 1 file changed, 264 insertions(+), 300 deletions(-)

diff --git a/clang/test/CodeGen/aarch64-fp8-intrinsics/acle_sme2_fp8_scale.c b/clang/test/CodeGen/aarch64-fp8-intrinsics/acle_sme2_fp8_scale.c
index b733e772ba3075..6bcf9bc946b20f 100644
--- a/clang/test/CodeGen/aarch64-fp8-intrinsics/acle_sme2_fp8_scale.c
+++ b/clang/test/CodeGen/aarch64-fp8-intrinsics/acle_sme2_fp8_scale.c
@@ -18,25 +18,21 @@
 // Single x2
 // CHECK-LABEL: @test_svscale_single_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv8f16(<vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]])
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svscale_single_f16_x213svfloat16x2_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x i16> [[OP2:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP6]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv8f16(<vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]]
 //
 svfloat16x2_t test_svscale_single_f16_x2(svfloat16x2_t op1, svint16_t op2) __arm_streaming
 {
@@ -45,25 +41,21 @@ svfloat16x2_t test_svscale_single_f16_x2(svfloat16x2_t op1, svint16_t op2) __arm
 
 // CHECK-LABEL: @test_svscale_single_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x i32> [[OP2:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv4f32(<vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]])
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svscale_single_f32_x213svfloat32x2_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x i32> [[OP2:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP6]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv4f32(<vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]]
 //
 svfloat32x2_t test_svscale_single_f32_x2(svfloat32x2_t op1, svint32_t op2) __arm_streaming
 {
@@ -72,25 +64,21 @@ svfloat32x2_t test_svscale_single_f32_x2(svfloat32x2_t op1, svint32_t op2) __arm
 
 // CHECK-LABEL: @test_svscale_single_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x i64> [[OP2:%.*]])
-// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv2f64(<vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]])
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svscale_single_f64_x213svfloat64x2_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x i64> [[OP2:%.*]])
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 0
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP3]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], 1
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP6]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x2.nxv2f64(<vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]]
 //
 svfloat64x2_t test_svscale_single_f64_x2(svfloat64x2_t op1, svint64_t op2) __arm_streaming
 {
@@ -100,37 +88,29 @@ svfloat64x2_t test_svscale_single_f64_x2(svfloat64x2_t op1, svint64_t op2) __arm
 // Single x4
 // CHECK-LABEL: @test_svscale_single_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[OP1_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[OP1_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv8f16(<vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], <vscale x 8 x i16> [[OP2:%.*]])
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svscale_single_f16_x413svfloat16x4_tu11__SVInt16_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x i16> [[OP2:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP12]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[OP1_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[OP1_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv8f16(<vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], <vscale x 8 x i16> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]]
 //
 svfloat16x4_t test_svscale_single_f16_x4(svfloat16x4_t op1, svint16_t op2) __arm_streaming
 {
@@ -139,37 +119,29 @@ svfloat16x4_t test_svscale_single_f16_x4(svfloat16x4_t op1, svint16_t op2) __arm
 
 // CHECK-LABEL: @test_svscale_single_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[OP1_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[OP1_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv4f32(<vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], <vscale x 4 x i32> [[OP2:%.*]])
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svscale_single_f32_x413svfloat32x4_tu11__SVInt32_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x i32> [[OP2:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP12]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[OP1_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[OP1_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv4f32(<vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], <vscale x 4 x i32> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]]
 //
 svfloat32x4_t test_svscale_single_f32_x4(svfloat32x4_t op1, svint32_t op2) __arm_streaming
 {
@@ -178,37 +150,29 @@ svfloat32x4_t test_svscale_single_f32_x4(svfloat32x4_t op1, svint32_t op2) __arm
 
 // CHECK-LABEL: @test_svscale_single_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[OP1_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[OP1_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv2f64(<vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], <vscale x 2 x i64> [[OP2:%.*]])
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z26test_svscale_single_f64_x413svfloat64x4_tu11__SVInt64_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i64> [[OP2:%.*]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 2
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 3
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP12]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[OP1_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[OP1_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.single.x4.nxv2f64(<vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], <vscale x 2 x i64> [[OP2:%.*]])
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]]
 //
 svfloat64x4_t test_svscale_single_f64_x4(svfloat64x4_t op1, svint64_t op2) __arm_streaming
 {
@@ -218,29 +182,29 @@ svfloat64x4_t test_svscale_single_f64_x4(svfloat64x4_t op1, svint64_t op2) __arm
 // Multi x2
 // CHECK-LABEL: @test_svscale_f16_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP2]], i64 8)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[OP2_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[OP2_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x2.nxv8f16(<vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svscale_f16_x213svfloat16x2_t11svint16x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv16i16(<vscale x 16 x i16> [[OP2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x2.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> poison, <vscale x 8 x half> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 16 x half> @llvm.vector.insert.nxv16f16.nxv8f16(<vscale x 16 x half> [[TMP6]], <vscale x 8 x half> [[TMP7]], i64 8)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x half> [[TMP8]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[OP2_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP2]], <vscale x 8 x i16> [[OP2_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x2.nxv8f16(<vscale x 8 x half> [[TMP4]], <vscale x 8 x half> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]]
 //
 svfloat16x2_t test_svscale_f16_x2(svfloat16x2_t op1, svint16x2_t op2) __arm_streaming
 {
@@ -249,29 +213,29 @@ svfloat16x2_t test_svscale_f16_x2(svfloat16x2_t op1, svint16x2_t op2) __arm_stre
 
 // CHECK-LABEL: @test_svscale_f32_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP2]], i64 4)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[OP2_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[OP2_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x2.nxv4f32(<vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svscale_f32_x213svfloat32x2_t11svint32x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv8f32(<vscale x 8 x float> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv8i32(<vscale x 8 x i32> [[OP2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x2.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> poison, <vscale x 4 x float> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 8 x float> @llvm.vector.insert.nxv8f32.nxv4f32(<vscale x 8 x float> [[TMP6]], <vscale x 4 x float> [[TMP7]], i64 4)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x float> [[TMP8]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[OP2_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP2]], <vscale x 4 x i32> [[OP2_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x2.nxv4f32(<vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]]
 //
 svfloat32x2_t test_svscale_f32_x2(svfloat32x2_t op1, svint32x2_t op2) __arm_streaming
 {
@@ -280,29 +244,29 @@ svfloat32x2_t test_svscale_f32_x2(svfloat32x2_t op1, svint32x2_t op2) __arm_stre
 
 // CHECK-LABEL: @test_svscale_f64_x2(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP2]], i64 2)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[OP2_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[OP2_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x2.nxv2f64(<vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svscale_f64_x213svfloat64x2_t11svint64x2_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[OP2]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x2.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]])
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 0
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP5]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP4]], 1
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP6]], <vscale x 2 x double> [[TMP7]], i64 2)
-// CPP-CHECK-NEXT:    ret <vscale x 4 x double> [[TMP8]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[OP2_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP2]], <vscale x 2 x i64> [[OP2_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x2.nxv2f64(<vscale x 2 x double> [[TMP4]], <vscale x 2 x double> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]]
 //
 svfloat64x2_t test_svscale_f64_x2(svfloat64x2_t op1, svint64x2_t op2) __arm_streaming
 {
@@ -312,45 +276,45 @@ svfloat64x2_t test_svscale_f64_x2(svfloat64x2_t op1, svint64x2_t op2) __arm_stre
 // Multi x4
 // CHECK-LABEL: @test_svscale_f16_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 16)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 24)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2]], i64 8)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2]], i64 16)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2]], i64 24)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[OP1_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[OP1_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[OP2_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], <vscale x 8 x i16> [[OP2_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]], <vscale x 8 x i16> [[OP2_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP6]], <vscale x 8 x i16> [[OP2_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 0
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 1
+// CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 2
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x4.nxv8f16(<vscale x 8 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], <vscale x 8 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], <vscale x 8 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], <vscale x 8 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]])
+// CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP16]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svscale_f16_x413svfloat16x4_t11svint16x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[OP1]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 8 x i16> @llvm.vector.extract.nxv8i16.nxv32i16(<vscale x 32 x i16> [[OP2]], i64 24)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x4.nxv8f16(<vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]], <vscale x 8 x i16> [[TMP4]], <vscale x 8 x i16> [[TMP5]], <vscale x 8 x i16> [[TMP6]], <vscale x 8 x i16> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> poison, <vscale x 8 x half> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP12]], <vscale x 8 x half> [[TMP13]], i64 16)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 32 x half> @llvm.vector.insert.nxv32f16.nxv8f16(<vscale x 32 x half> [[TMP14]], <vscale x 8 x half> [[TMP15]], i64 24)
-// CPP-CHECK-NEXT:    ret <vscale x 32 x half> [[TMP16]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } poison, <vscale x 8 x half> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP0]], <vscale x 8 x half> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP1]], <vscale x 8 x half> [[OP1_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP2]], <vscale x 8 x half> [[OP1_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } poison, <vscale x 8 x i16> [[OP2_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP4]], <vscale x 8 x i16> [[OP2_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP5]], <vscale x 8 x i16> [[OP2_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = insertvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP6]], <vscale x 8 x i16> [[OP2_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 0
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 1
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 2
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP7]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } @llvm.aarch64.sme.fp8.scale.x4.nxv8f16(<vscale x 8 x half> [[TMP8]], <vscale x 8 x half> [[TMP9]], <vscale x 8 x half> [[TMP10]], <vscale x 8 x half> [[TMP11]], <vscale x 8 x i16> [[TMP12]], <vscale x 8 x i16> [[TMP13]], <vscale x 8 x i16> [[TMP14]], <vscale x 8 x i16> [[TMP15]])
+// CPP-CHECK-NEXT:    ret { <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half> } [[TMP16]]
 //
 svfloat16x4_t test_svscale_f16_x4(svfloat16x4_t op1, svint16x4_t op2) __arm_streaming
 {
@@ -359,45 +323,45 @@ svfloat16x4_t test_svscale_f16_x4(svfloat16x4_t op1, svint16x4_t op2) __arm_stre
 
 // CHECK-LABEL: @test_svscale_f32_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 8)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 12)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2]], i64 4)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2]], i64 8)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2]], i64 12)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[OP1_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[OP1_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[OP2_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], <vscale x 4 x i32> [[OP2_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]], <vscale x 4 x i32> [[OP2_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP6]], <vscale x 4 x i32> [[OP2_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 0
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 1
+// CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 2
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x4.nxv4f32(<vscale x 4 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], <vscale x 4 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], <vscale x 4 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], <vscale x 4 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]])
+// CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP16]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svscale_f32_x413svfloat32x4_t11svint32x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 4 x float> @llvm.vector.extract.nxv4f32.nxv16f32(<vscale x 16 x float> [[OP1]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.extract.nxv4i32.nxv16i32(<vscale x 16 x i32> [[OP2]], i64 12)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x4.nxv4f32(<vscale x 4 x float> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[TMP2]], <vscale x 4 x float> [[TMP3]], <vscale x 4 x i32> [[TMP4]], <vscale x 4 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP6]], <vscale x 4 x i32> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> poison, <vscale x 4 x float> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP12]], <vscale x 4 x float> [[TMP13]], i64 8)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 16 x float> @llvm.vector.insert.nxv16f32.nxv4f32(<vscale x 16 x float> [[TMP14]], <vscale x 4 x float> [[TMP15]], i64 12)
-// CPP-CHECK-NEXT:    ret <vscale x 16 x float> [[TMP16]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } poison, <vscale x 4 x float> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP0]], <vscale x 4 x float> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP1]], <vscale x 4 x float> [[OP1_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP2]], <vscale x 4 x float> [[OP1_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } poison, <vscale x 4 x i32> [[OP2_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP4]], <vscale x 4 x i32> [[OP2_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP5]], <vscale x 4 x i32> [[OP2_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = insertvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP6]], <vscale x 4 x i32> [[OP2_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 0
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 1
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 2
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP7]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sme.fp8.scale.x4.nxv4f32(<vscale x 4 x float> [[TMP8]], <vscale x 4 x float> [[TMP9]], <vscale x 4 x float> [[TMP10]], <vscale x 4 x float> [[TMP11]], <vscale x 4 x i32> [[TMP12]], <vscale x 4 x i32> [[TMP13]], <vscale x 4 x i32> [[TMP14]], <vscale x 4 x i32> [[TMP15]])
+// CPP-CHECK-NEXT:    ret { <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float> } [[TMP16]]
 //
 svfloat32x4_t test_svscale_f32_x4(svfloat32x4_t op1, svint32x4_t op2) __arm_streaming
 {
@@ -406,45 +370,45 @@ svfloat32x4_t test_svscale_f32_x4(svfloat32x4_t op1, svint32x4_t op2) __arm_stre
 
 // CHECK-LABEL: @test_svscale_f64_x4(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 2)
-// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 4)
-// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 6)
-// CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2:%.*]], i64 0)
-// CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2]], i64 2)
-// CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2]], i64 4)
-// CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2]], i64 6)
-// CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[OP1_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[OP1_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP4:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[OP2_COERCE0:%.*]], 0
+// CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], <vscale x 2 x i64> [[OP2_COERCE1:%.*]], 1
+// CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]], <vscale x 2 x i64> [[OP2_COERCE2:%.*]], 2
+// CHECK-NEXT:    [[TMP7:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP6]], <vscale x 2 x i64> [[OP2_COERCE3:%.*]], 3
+// CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 3
+// CHECK-NEXT:    [[TMP12:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 0
+// CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 1
+// CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 2
+// CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 3
+// CHECK-NEXT:    [[TMP16:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x4.nxv2f64(<vscale x 2 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], <vscale x 2 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], <vscale x 2 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], <vscale x 2 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]])
+// CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP16]]
 //
 // CPP-CHECK-LABEL: @_Z19test_svscale_f64_x413svfloat64x4_t11svint64x4_t(
 // CPP-CHECK-NEXT:  entry:
-// CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv8f64(<vscale x 8 x double> [[OP1]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP4:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2:%.*]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP5:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP6:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP7:%.*]] = tail call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[OP2]], i64 6)
-// CPP-CHECK-NEXT:    [[TMP8:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x4.nxv2f64(<vscale x 2 x double> [[TMP0]], <vscale x 2 x double> [[TMP1]], <vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]])
-// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 0
-// CPP-CHECK-NEXT:    [[TMP10:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> poison, <vscale x 2 x double> [[TMP9]], i64 0)
-// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 1
-// CPP-CHECK-NEXT:    [[TMP12:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], i64 2)
-// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 2
-// CPP-CHECK-NEXT:    [[TMP14:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP12]], <vscale x 2 x double> [[TMP13]], i64 4)
-// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP8]], 3
-// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call <vscale x 8 x double> @llvm.vector.insert.nxv8f64.nxv2f64(<vscale x 8 x double> [[TMP14]], <vscale x 2 x double> [[TMP15]], i64 6)
-// CPP-CHECK-NEXT:    ret <vscale x 8 x double> [[TMP16]]
+// CPP-CHECK-NEXT:    [[TMP0:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } poison, <vscale x 2 x double> [[OP1_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP1:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP0]], <vscale x 2 x double> [[OP1_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP2:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP1]], <vscale x 2 x double> [[OP1_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP3:%.*]] = insertvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP2]], <vscale x 2 x double> [[OP1_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP4:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } poison, <vscale x 2 x i64> [[OP2_COERCE0:%.*]], 0
+// CPP-CHECK-NEXT:    [[TMP5:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP4]], <vscale x 2 x i64> [[OP2_COERCE1:%.*]], 1
+// CPP-CHECK-NEXT:    [[TMP6:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP5]], <vscale x 2 x i64> [[OP2_COERCE2:%.*]], 2
+// CPP-CHECK-NEXT:    [[TMP7:%.*]] = insertvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP6]], <vscale x 2 x i64> [[OP2_COERCE3:%.*]], 3
+// CPP-CHECK-NEXT:    [[TMP8:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 0
+// CPP-CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 1
+// CPP-CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 2
+// CPP-CHECK-NEXT:    [[TMP11:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP3]], 3
+// CPP-CHECK-NEXT:    [[TMP12:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 0
+// CPP-CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 1
+// CPP-CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 2
+// CPP-CHECK-NEXT:    [[TMP15:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP7]], 3
+// CPP-CHECK-NEXT:    [[TMP16:%.*]] = tail call { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sme.fp8.scale.x4.nxv2f64(<vscale x 2 x double> [[TMP8]], <vscale x 2 x double> [[TMP9]], <vscale x 2 x double> [[TMP10]], <vscale x 2 x double> [[TMP11]], <vscale x 2 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP13]], <vscale x 2 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]])
+// CPP-CHECK-NEXT:    ret { <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double> } [[TMP16]]
 //
 svfloat64x4_t test_svscale_f64_x4(svfloat64x4_t op1, svint64x4_t op2) __arm_streaming
 {